Swift Transformers Reaches 1.0 – and Looks to the Future

swift-transformers was launched two years ago with the goal to support Apple developers and help them integrate local LLMs in their apps. A lot has changed since then (MLX and chat templates did not exist!), and we’ve learned how the community is actually using the library. We want to double down on the use cases that provide most benefits to the community, and lay out the foundations for the future. After this release, we’ll focus a lot on MLX and agentic use cases.

swift-transformers is a Swift library that aims to reduce the friction for developers that want to work with local models on Apple Silicon platforms, including iPhones. It includes the missing pieces that are not provided by Core ML or MLX alone, but that are required to work with local inference. Namely, it provides the following components:

• Tokenizers: Preparing inputs for a language model is surprisingly complex. We've built a lot of experience with our Python and Rust libraries, and we wanted to bring the same performant, ergonomic experience to Swift.
• Hub: This is an interface to the Hugging Face Hub, where all open models are available. It allows you to download models from the Hub and cache them locally, supporting background resumable downloads and offline mode.
• Models and Generation: These are wrappers for LLMs converted to the Core ML format, making it easy to run inference with them.

Notable projects that rely on swift-transformers include Apple's mlx-swift-examples, WhisperKit by argmax, and FastVLM.

Version 1.0 signals stability in the package. Key updates include:

• Modularization: Tokenizers and Hub are now first-class, top-level modules.
• New swift-jinja: A collaboration to create a faster chat template library.
• Modern Core ML APIs: Support for stateful models (for easier KV-caching) and expressive MLTensor APIs.
• Swift 6 Support: Fully supported for modern development.

Looking forward, we are super interested in exploring MLX further and are extremely excited about agentic use in general and MCP (Model Context Protocol) in particular. We think that exposure of system resources to local workflows would be a game changer.